40 IdeaBook 2015
collected simply for documentation and then are stored
to comply with regulatory statutes. In other cases it may
be because analysis takes place on a descriptive level
only, and no action is taken as a result of the findings.
This latter situation is all too common; many companies
have elaborately designed dashboards that
present useful information but are not
backed up by processes and workflows
that translate those findings into actions.
Many optimization problems in supply
chain management represent instances
where a closer look at available data can
improve decision making. One example is lot sizing in production planning.
Although most companies retain ordering
data in structured form, “educated guessing” guides many production dispatchers’
day-to-day routines. This practice leads to
wasted resources, especially when inputs
are perishable and material and setup
costs are high, as they are in the pharmaceutical industry.
Today many optimization routines in enterprise
resource planning (ERP) systems are still based on variations of the classic linear optimization method or upon
the heuristics derived from it. Despite their advantages,
models applying linear optimization bear a considerable
disadvantage that makes them unsuitable for many
real-life scenarios—their specification requires information that may be fragmented or just not available at
the required level of detail. In addition, the underlying
data structure may shift dramatically without the model
accounting for the new pattern.
Such challenges can be met by applying nonparamet-ric algorithms to the evaluation problem. Mother Nature
provides an example of such an algorithm: “Survival of
the fittest” is a robust method for solving adaptation
(that is, optimization) challenges. “Genetic algorithms”
mimic the evolutionary process and natural selection.
Applied to the lot-sizing problem, a genetic algorithm
exchanges permutations of orders, which is analogous to
the permutation of chromosomes in the natural process
of reproduction. The results are compared against a target function, and well-fitting permutations are chosen
for new iterations. This method has clear advantages.
In a research project, when we compared the results
obtained with this algorithm to the current state of
human decision making at a contract pharmaceutical
manufacturer, we identified a savings potential of up
to 8. 7 percent of the total purchasing costs of the prod-
uct group where the algorithm was applied. Moreover,
the genetic algorithm outperformed the classic linear
modeling approach by 26 percent. Our findings support
the opinion of many experts that existing
internal data often hold sufficient infor-
mation to optimize networks.
Still, sometimes it is worthwhile to
think outside the box. Simply combining
separate data domains can yield exciting
insights. For instance, by juxtaposing its
point-of-sale data with severe weather
warnings, Wal-Mart discovered a remarkable pattern: In areas threatened by hurricanes, not only did the demand for
emergency-relief equipment increase, but
people also hoarded Pop-Tarts (a sweet
breakfast pastry sold in North America).
This nonintuitive finding was produced
by a simple correlation analysis; it now
helps the retailer ensure that regions facing a potential
natural disaster have sufficient supplies of water, shovels—and Pop-Tarts.
7
There are cases, however, where such retrospective
insights are not sufficient to meet the challenges of a particular environment. Real-time data will then be needed
to transform supply chains into dynamically adapting
networks. In that case there is no choice but to seek new
data sources.
A NEW AGE OF REAL-TIME VISIBILITY
By tapping sources like Web search queries and social
media in addition to data provided by sensors and
mobile devices, companies gain access to a massive and
steadily growing flow of data. This flow, characterized
by its unprecedented volume, velocity, and variety, is
known as “big data.” The potential economic implications of big data are huge; for instance, the consulting
firm McKinsey & Company expects the U.S. health care
industry alone to create US $300 billion in value by using
big data to drive efficiency and quality.
8
The use of unstructured data sources, such as Web
search queries, has proved especially useful in increasing
the accuracy of predictions for outcomes of events in
the immediate future. For example, it has been demonstrated that Web search queries can predict influenza
epidemics9 or the commercial success of movies, music,
